不動產交易在任何國家都是一項重要的經濟活動,其單筆交易金額是鉅大的,學者 及業界人士都積極投入研究,買賣雙方對於大筆金額交易也會特別在意價格 關於 價格預測通常以回歸分析 方法來做連續型資料的預測 。 政府於2012 年實行不動產交易實價登錄制度,提供公開且免費的交易資料,因此 本研究以台中市 2 016~2019 年的 實價登錄資料為基礎但由於資料中的地址欄位 並 非精準揭露,因此使用地理資訊圖資雲服務平台(TGOS Taiwan Geospatial One-Stop) ,將模糊性的區段位置轉換成較精確性的地理坐標資訊再將其他資料做資料預處理(Data preprocessing)以利建立模型分析,本研究應用最小二乘支持向量回歸LSSVR進行不動產價格 預測,並與其他機器學習模型及多元線性回歸相比較,實驗結果 最小二乘支持向量回歸LSSVR 皆優於倒傳遞神經網路BPNN 、廣義回歸神經網路GRNN 、分類及回歸樹CART 及多元線性回歸 MLR 。
Real estate transaction is an important economic activity in many countries and has been of great research interest to scholars and people in the industry because of the huge amount of money involved. In 2012, the government implemented the real-value transaction registration system for real estate transactions to provide open and free transaction information. This study is based on the actual price registration data in Taichung City from 2016 to 2019. Because the address information of the data is not accurately disclosed, the Geographic Information Map Cloud Service platform (TGOS, Taiwan Geospatial One-Stop) is used to convert the ambiguous locations into more accurate geographic coordinate information, and other data are preprocessed so as to establish a model analysis. In this study, I use least squares support vector regression (LSSVR) to predict real estate prices. The results of the experiment show that the prices predicted by LSSVR are more accurate than those predicted by back propagation neural network (BPNN), generalized regression neural network (GRNN), classification and regression tree (CART) and multiple linear regression (MLR).